One paper I'm thinking of compared implementations of two different algorithms, and showed that for chemical data VF2 was several times faster than the Ullman algorithm for subgraph isomorphism. A couple of years earlier, two of the free software projects in the field did the same analysis, with the same conclusion. Both published their results on their respective blogs/wiki, and changed the internals to use VF2.
These don't resemble a publication in a way that Google Scholar can discern.
It's a bit annoying to me in that the scientific literature is supposed to be "self-correcting", in the sense that I could publish a followup paper highlighting some of the pre-history. But the journal I'm thinking of is an OA journal, with no letters to the editor or similar section. The only way to update the literature is to pay ~$1,000 for a full-sized paper, or convince some other journal (... or arxiv? Hmm...) to publish a correction piece.
Quite annoying.
[1] https://scholar.google.fr/scholar?q=rauzy+ARMv7-M